1 A SAR speckle filter based on Residual Convolutional Neural Networks Alessandro Sebastianelli ID , Member, IEEE * , Maria Pia Del Rosso ID , Member, IEEE * , Silvia L. Ullo ID , Senior Member, IEEE * Abstract—In recent years, Machine Learning (ML) algorithms have become widespread in all fields of Remote Sensing (RS) and Earth Observation (EO). This has allowed a rapid development of new procedures to solve problems affecting these sectors. In this context, the authors of this work aim to present a novel method for filtering the speckle noise from Sentinel-1 data by applying Deep Learning (DL) algorithms, based on Convolutional Neural Networks (CNNs). The obtained results, if compared with the state of the art, show a clear improvement in terms of Peak Signal-to-Noise Ratio (PSNR) and Structural Similarity Index (SSIM), by proving the effectiveness of the proposed architecture. Moreover, the generated open-source code and dataset have been made available for further developments and investigation by interested researchers. Index Terms—Synthetic Aperture Radar (SAR), Sentinel-1, noise filtering, speckle filtering, Artificial Intelligence (AI), Deep Learning (DL), Convolutional Neural Networks (CNNs). I. I NTRODUCTION As well known, Synthetic Aperture Radar (SAR) data are affected by several types of noise, and among them, the speckle noise, resulting into a positive or negative interference phenomenon that occurs in each resolution cell. Speckle confers to SAR data a granular aspect, resulting into a high degradation of the image quality [1], unlike optical data, that are characterized by clear and recognizable features. An example of two different images based on SAR data (with speckle) and optical data is shown in Figure 1, where it is evident how the speckle presence prejudices the visual analysis and interpretation of the Area of Interest (AOI). Moreover, speckle also affects the phase of scattering coefficients for the analyzed surface and degrades the signal polarimetric information [2]. Speckle results in misclassification errors and in many other problems if SAR data are not correctly pre-processed. Classical speckle filtering techniques can be divided into two macro categories: 1) single product speckle filtering, 2) multi-temporal speckle filtering. Single product speckle filtering methods are mainly based on mathematical models of the speckle phenomenon, and a classic approach in this category is the multilooking, a process that can reduce the speckle by means of incoherent averag- ing of many sub-aperture acquisitions (looks). This process normally includes the sub-sampling of data and the ground- projection, leading to the increase of the radiometric resolution * The authors are with the Departement of Engineering, University of Sannio, Benevento, Italy. A. Sebastianelli (email: sebastianelli@unisannio.it), M. P. Del Rosso (email: mpdelrosso@unisannio.it) and S. L. Ullo (email: ullo@unisannio.it). (a) Sentinel-1A acquisition on 2021-01-01 focusing on Rome (Italy), Interferometric Wide Swat (IW) mode, Single Look Complex (SLC) data, VV polarization (Intensity). (b) Sentinel-2A acquisition on 2021-01-01 focusing on Rome (Italy), B4, B3 and B2 bands (RGB channels). Fig. 1: Comparison between Sentinel-1 (SAR) and Sentinel-2 (optical) acquisitions. and the decrease of the spatial one. Yet, after this process the phase information is lost [3]. Another typical approach in the single product speckle filtering category involves the use arXiv:2104.09350v1 [cs.AI] 19 Apr 2021